Prediction in traditional, reductionist natural science serves the role of validating hypotheses about invariant natural phenomena. In recent years, a new type of prediction has arisen in science, motivated in part by the needs of policy makers and the availability of new technologies. This new predictive science seeks to foretell the behavior of complex environmental phenomena such as climate change, earthquakes, and extreme weather events. Significant intellectual and financial resources are now devoted to such efforts, in the expectation that predictions will guide policy making. These expectations, however, derive in part from confusion about the different roles of prediction in science and society. Policy makers lack a framework for assessing when and if prediction can help achieve policy goals. This project is a first step toward developing such a framework.